7 stories in this blend

Meta's AI assistant Muse is instructed to maintain and hourly update detailed profiles on people in a user's network. The tracking occurs automatically without the affected individuals opting in or giving consent.

Startup Kled AI updated its platform to gather training data using physical and digital human demonstrations. Over half a million contributors supply video, text, image, and audio clips that allow AI developers to order custom training datasets delivered within three days.

A new company called HANDO Labs is building a crowdsourced database by rewarding people with stock exposure for recording daily tasks. Participants submit videos of activities like object sorting to assist robotics teams in training hardware models.

Robotics firm SOMA introduced soma one, a mobile robot equipped with two arms and cameras designed to record human task demonstrations. The system includes virtual reality controls and outputs motion data formatted directly for machine learning pipelines. It offers academic and commercial laboratories a ready made hardware platform for testing physical AI capabilities.

SpaceX has engaged in negotiations to buy training datasets accumulated by failed technology startups. The hardware firm plans to use these assets to boost its proprietary AI development efforts.

An investigation found that organizations are acquiring uncommon physical books to scan them into AI training datasets before shredding the original paper copies. Tracking devices planted inside targeted books helped trace the destroyed volumes back to facilities linked to major tech companies.

Google successfully bid $10 million in a bankruptcy court auction to buy internal corporate messages from Spirit Airlines. The purchased dataset contains 100 million corporate emails along with 500 million chat logs, though personal customer information was excluded from the sale.